New directions (Ideas)

New datasets to consider

Using Internet speed as a proxy for a region’s development

• Use Internet speeds for a location to estimate the development in the area (Ookla provides open source datasets for the same)

Using Openstreetmap data and its attributes

• For a given location / area using Openstreet Map data it is possible to derive the following information, which can be used to estimate wealth
  • No. of buildings
  • No. of roads
  • No. of primary roads
  • No. of trunk roads
  • count of market places
  • Number of charging stations
  • No. of post offices
  • Supermarket counts
  • Car repair counts
  • Department stores
  • Computers
  • Playgrounds
  • Monuments

New Modelling approaches to consider

GAM Models

• Using explainable models - How much a variable is influencing the predictions
• Explainability along with good prediction accuracy. Example - Explainable boosting machines (EBMs)
• We can build editable models using EBMs

Using Semi-supervised learning

• Use libraries like snorkel to generate labels
• Low confidence images can be routed to humans for labeling

Using Bayesian Updates

• There is correlation between wealth and spatial location
• A Gaussian process can be implemented on top of the model. Use the prior observations to get the posterior distribution from the priors (tried at stanford in 2016)

Time series Analysis

• Using time series analysis to forecast the wealth from past surveys